Abstract:
Seasonal Climate Forecasting (SCF) in South Africa has a history spanning several decades. During this period a number of SCF systems have been developed for the prediction of seasonal-to-interannual variability of rainfall and surface temperatures. Areas of highest predictability, albeit relatively modest, have also been identified. The north-eastern parts of South Africa that includes the Limpopo province has been demonstrated to be one of the areas of highest SCF skill in the country. Statistical post-processing techniques applied to global climate model output were part of this forecast system development, and were subsequently successfully used in the construction of forecasts systems for applications in sectors which are associated with ENSO-driven climate variability, such as dry-land crop yields and river flows. Here we follow a similar post-model processing approach to test SCF systems for application to the incidence of seasonal malaria in Limpopo. The malaria forecast system introduced here makes use of the seasonal rainfall output fields of one of the North American Multi-Model Ensemble (NMME) climate models, which is then linked statistically through multiple linear regression to observed malaria incidence. The verification results as calculated over a 20-year hindcast period show that the season of highest malaria incidence forecast skill is during the austral mid-summer time of December to February. Moreover, the hindcasts based on the NMME model outscore those of statistical forecast models that separately use Indian and Pacific Ocean sea-surface temperatures as predictors, thus justifying the use of physical global climate models for this kind of application. Additional results indicate that model skill levels may include quasi-decadal variability, that the periods over which forecast verification is performed strongly influences forecast skill, and that poorly predicted malaria seasons may have serious financial implications on public health operations.
